Collection of experiments that can be used for explainable AI
| Original image | Input x Gradient |
|---|---|
| TODO | TODO |
As implemented in https://arxiv.org/abs/1412.6572
Adversial attack aims to introduce a noice which will cause our prediction to go wrong.
| Original Image | Adversial Attack |
|---|---|
 |
 |
As implemented in https://arxiv.org/pdf/1910.01279.pdf
Scorecam aims to make a hypothesis what region of the image our network will make its decision based on. Whiter the pixels are in the picture, more attention the model pays to the region.
| Original Image | Scorecam Focus |
|---|---|
 |
 |